>> On the science/biostatistics side, do you know of any company that has been able to do an interim look without any p-value penalty? If GNVC were able to get the FDA to change the protocol to OS survival rather than 12-month survival, then how do you see this affecting the enrollment numbers and the timelines? <<
I think the first interim look by GNVC was near penalty-free , since I suspect they only allocated a tiny p-value , knowing there was no chance for statsig so soon. For any look where you hope to get trial-stopping results , the allocation will be meaningful and will reduce remaining alpha , though maybe not in a strictly additive way , as Clark points out. There's really no avoiding this , otherwise everyone would take interim looks all the time.
I think the chances of getting P=0.05 at 12 months are probably similar by chi-square analysis of proportions surviving or by a log-rank OS analysis thru 12 months. If not , I doubt the FDA would allow the choice between methods. I think FDA prefers log-rank OS because it's more robust and less susceptible to chance variations that might result in a false-positive. That's why I suggested that maybe GNVC would have some bargaining power by agreeing to switch to OS. Then , as I see it anyway , the critical thing is whether the survival advantage of TNFerade continues growing at times past 12 months. If it does , they might get to statsig much faster by being able to include data out to 18 months or beyond. It then just becomes a matter of how many events are necessary , and what alpha they choose for the interim look. I have no idea what those figures might be , but I'd guess they'd enroll at least half of the original 300 , and let the last ones simmer for a year or so before the next look , using a P~0.01 to 0.02. I think they'd continue to enroll until they see the results of the second look , just to be safe.
Maybe some of the stat gurus can run a few thousand simulations for us and give us a better estimate. I'd ask , but I've been hard on the statisticians lately. :)
AI
